effect of HT LINK SPEED on performance - myth or reality???

[Urizen]

I CHANGED IT BACK TO 800 AND GOT THE SAME PERFORMANCE

*phew* <- mega-sigh of relief

THANKS A LOT M8 that's a relief


to all the others who told me not to worry about HT link : forgive me for having doubted yall

Another Question Will The Htlink Make The Cpu Run Hotter?

nope, only global frequency counts (baseline bus frequency (200mhz by default) X cpu multiplier)

 

[Mussels]

Another Question Will The Htlink Make The Cpu Run Hotter?

no. Increasing MHz rarely increases heat, its voltage that does that.

(yes it increases a little, but its a night and day difference compared to voltage)

 

[xvi]

It should make your chipset run hotter too. Again, not by much.

 

[Urizen]

but on the quad core phenoms expect a 20% difference.

ah so 1000ghz HT link speeds is still required for quad cores ?


I ask this 'cause my mobo supports AM2+ CPUs as well (phenoms) so you're saying that a quad core CPU would be overkill for my mobo as it wouldn't be running at its full potential ? (since my mobo only does HT up to 800mhz)



so to summarize :

800Mhz HT is enough for 1 or 2 core CPUs
but
800Mhz is not enough for 4 core CPUs ?

is that correct ?

 

[xvi]

I just tested a dual proc, dual core Opteron 1U Gateway server with 1000MHz HT. Prime95 loaded the HT to about 700 MAX.

A Phenom isn't going to be much different.

MYTH, people...

 

[hat]

Yeah, I didn't get a difference in SuperPi with my HT speed running at 2500MT/s vs 2000MT/s... well, not enough to change the score by even one second (at that time I was using the non-modifed super-pi that only gave you whole-number values, no decimels).

 

[Urizen]

I just tested a dual proc, dual core Opteron 1U Gateway server with 1000MHz HT. Prime95 loaded the HT to about 700 MAX.

A Phenom isn't going to be much different.

MYTH, people...

wait just so we're clear : you mean dual proc AND dual core ? so a total of 4 cores ?

Yeah, I didn't get a difference in SuperPi with my HT speed running at 2500MT/s vs 2000MT/s... well, not enough to change the score by even one second (at that time I was using the non-modifed super-pi that only gave you whole-number values, no decimels).

but what's 2000MT/s - is that 1000mhz ?

if so did you also test it at 1600MT/s for comparison with 2000MT/s? that's the important part

 

[xvi]

Thank you, hat. Point and case.

 

[xvi]

wait just so we're clear : you mean dual proc AND dual core ? so a total of 4 cores ?

but what's 2000MT/s - is that 1000mhz ?

They added DDR to the HyperTransport bus. AMD should be rolling out 3600MT/s HyperTransport soon.

 

[Urizen]

They added DDR to the HyperTransport bus.

ok so that means he only tested it at 1000Mhz and 1250mhz and the results were the same, but he didn't test it at 800Mhz (1600MT/s) which is my mobo's limit

hat> what are the results @ 1600MT/s?

 

[hat]

By 2000MT/s I mean 1000MHz HTT link. 2500MT/s is 1250MHz, and right now at 2420MT/s that is 1210MHz... get it?

MT/s means Mega Transfers/second if you want to know. I don't know what Mega Transfers means, but I get to do up to 2420 of them every second... at least you can look it up now that you know what you're looking for if you're that interested in it.

 

[Urizen]

ok thx so that means you tested at 1000Mhz and 1250Mhz and u said there was no difference by even 1 second

but what about 800Mhz ? (that's my mobo's limit) did you test it at 800Mhz ?
was it slower than with 1000mhz ? (I hope not )

 

[xvi]

..and yes. A total of four cores.

It's not going to be slower unless you have something else saturating the last 400MT/s of the HT bus like... 50 SATA Barracuda drives in RAID?

 

[hat]

ok thx so that means you tested at 1000Mhz and 1250Mhz and u said there was no difference by even 1 second

but what about 800Mhz ? (that's my mobo's limit) did you test it at 800Mhz ?
was it slower than with 1000mhz ? (I hope not )

No I did not test it at 800MHz. Let's see, right now my FSB is 242... so if I dropped the HT multi to 4x it would be 968MHz, or 1936MT/s. If I dropped it to 3x it would be 726, or 1452MT/s, 2x would be 484 or 968MT/s, and if I put it at 1x (I'm not even sure I have this setting but I think I do) it would be 242MHz or 484MT/s. Tell ya what...

I will run SuperPI 1m 3 times under the same circumstances with the MT/s speed at:
2420MT/s
1936MT/s
1452MT/s
968MT/s
484MT/s (if I have it).

I'll post back in about 5 min.

 

[Urizen]

k thx for the trouble I'll be waiting

 

[xvi]

..if you can even boot at 2x or 3x.

 

[Urizen]

I can boot at 3x 2x or even 1x

why, is lower HT speed less stable ?

 

[hat]

Here are my results running SuperPI mod 1.5 calculating 1M:
2420MT/s: 29.578s
1936MT/s: 29.547s
1452MT/s: 29.532s
968MT/s: 29.531s
484MT/s: 30.141s

Interesting indeed. It ran faster and faster until I booted at 1x and ran it. My conclusuion is that it's perfectly fine until it gets lower than your RAM speed, then it's slightly slower. My RAM, if you check my specs, is running at DDR645 4-4-4-12-1T.

My guess for it running faster on a lower HTT speed is purely out of luck of the small core frequency fluctuations. But it's kind of weird how consistient it is... maybe it fluctuates less at a lowet HTT speed.
Is 1m good enough for you or do you want me to run a longer test? My options are:
16K
32K
64K
128K
256K
512K
1M (what I did just now)
2M
4M
8M
16M
32M

 

[Mussels]

I can boot at 3x 2x or even 1x

why, is lower HT speed less stable ?

none of my 939 systems wouldnt boot at 3x, and my AM2's dont work below 5x.

Probably mobo dependant.

 

[Urizen]

Here are my results running SuperPI mod 1.5 calculating 1M:
2420MT/s: 29.578s
1936MT/s: 29.547s
1452MT/s: 29.532s
968MT/s: 29.531s
484MT/s: 30.141s

Interesting indeed. It ran faster and faster until I booted at 1x and ran it. My conclusuion is that it's perfectly fine until it gets lower than your RAM speed, then it's slightly slower. My RAM, if you check my specs, is running at DDR645 4-4-4-12-1T.

My guess for it running faster on a lower HTT speed is purely out of luck of the small core frequency fluctuations. But it's kind of weird how consistient it is... maybe it fluctuates less at a lowet HTT speed.
Is 1m good enough for you or do you want me to run a longer test? My options are:
16K
32K
64K
128K
256K
512K
1M (what I did just now)
2M
4M
8M
16M
32M

nah thanx a lot hat that's will do!

so I guess I can move on to a 4core on my HT800 mobo without losing any of its potential then, right ?
if so I'll get a Phenom as soon as they fix that L3 bug

none of my 939 systems wouldnt boot at 3x, and my AM2's dont work below 5x.

Probably mobo dependant.

that's weird

 

[hat]

My mobo works will all settings...
which is werid. Why would a motherboard have a setting available that it simply won't work with? And why would it not work at a SLOWER, EASIER TO RUN speed anyway?

 

[hat]

nah thanx a lot hat that's will do!

so I guess I can move on to a 4core on my HT800 mobo without losing any of its potential then, right ?
if so I'll get a Phenom as soon as they fix that L3 bug

that's weird

You seem new here. There is a thanks button under every post. If someone helped you, you can click the thanks button under thier helpful post and give them some extra stars/stripes/ribbons/medals to flaunt around on thier, uhh, shirt... I guess...

 

[Mussels]

My mobo works will all settings...
which is werid. Why would a motherboard have a setting available that it simply won't work with? And why would it not work at a SLOWER, EASIER TO RUN speed anyway?

well for all i know it was CPU dependant, and none of mine worked with it. dont have any of them around anymore, so i cant test.

Good info on those HT benchies - would you mind dropping your ram speed (leave the rest the same) and get it below 484 - and then do the tests again.
Would like to confirm your theory on HT should be above ram MHz

 

[Urizen]

You seem new here. There is a thanks button under every post. If someone helped you, you can click the thanks button under thier helpful post and give them some extra stars/stripes/ribbons/medals to flaunt around on thier, uhh, shirt... I guess...

yeah I know in fact you're not the 1st as I've thanked quite a lot of other ppl too on this thread (all those who wrote "optimistic" posts in fact )


cool sig u got btw - I think that's the 1st tech vB forum I've been to without censorship ^_^

 

[hat]

I'm running at 322.6MHz right now on a 266MHz divider. If I drop it to the 200MHz divider it will be running 242MHz. Which is DDR484, and equal to the speed of 484MT/s, lol...

but yeah I'll drop it and post results, but then since the RAM would be so slow I think the scores would be much more RAM influenced than HTT speed influenced. But I'll do it anyway..

As for the non-sensored sig, it wouldn't suprise me if Thermo came along and edited it for me, but even now the mods are pretty lax on things like that unless someone comes along, gets thier panties in a twist and reports it. HAT HAS A BAD WORD IN HIS SIG!!